H13A-1314
An Efficient and Effective Strategy in Uncertainty Evaluation of Conceptual Hydrologic Models
Two elementary issues in contemporary earth system science are (1) the specification of model parameters which characterize a system, (2) the estimation of dynamic state (prognostic) variables which express the system dynamic. Reliable estimation of these elements is needed to enable the model to generate the forecasts as accurate as possible. Emerging technologies in Bayesian estimation within the Monte Carlo framework provides a platform for improved estimation of hydrologic model components and uncertainty assessment by complete representation of forecast and analysis probability distributions. In this study, the major effort goes into introducing the recursive Bayesian information fusion technique within the context of stochastic filtering as an alternative approach to batch calibration to characterize and reduce the uncertainties associated with hydrologic model parameters and state variables. The issues of sequential sampling/resampling and also the sensitivity of the model performance to the resampling and ensemble size as two key components in this scheme are emphasized. The power and effectiveness of the procedures are demonstrated by streamflow data assimilation into a conceptual hydrologic model where predictive uncertainty as the final product of the methodology is obtained.
H13A-1315
Bayesian Model Averaging on Parameterization Non-Uniqueness and Conditional Uncertainty Analysis
Understanding of subsurface heterogeneity, e.g., hydraulic conductivity, is inherently difficult in that natural heterogeneity and processes are extremely complex and the available data are limited. Although the parameter structure error in groundwater modeling has been assessed with one parameterization method (zonation or interpolation), with limited information many parameterization methods may interpret the same data satisfactorily. To cope with the non-Uniqueness problem of parameterization, we introduce a Bayesian model averaging (BMA) method to integrate multiple parameterization methods in a Bayesian geostatistical framework. Moreover, a generalized parameterization (GP) method is adopted to estimate the highly complex spatial distribution of parameter heterogeneity. In this study, we combine BMA and GP as a Bayesian multi-parameterization (BMP) method to better represent the heterogeneity and reduce the model prediction uncertainty. The BMP avoids over-confidence in a single parameterization method. The proposed methodology is conducted in a numerical example where the spatially distributed hydraulic conductivity is estimated. The optimal weighting coefficients embedded in GP are identified through the maximum likelihood estimation (MLE) where the misfits between the observed and calculated groundwater heads are minimized. The conditional means and conditional covariances of the estimated hydraulic conductivity distribution are obtained to assess the estimation uncertainty.
H13A-1316
Assessing Recharge Model Uncertainty: Case Study Using the Death Valley Regional Flow System Model
Hydrologic analyses are commonly based on a single conceptual-mathematical model. Yet hydrologic environments are open and complex, rendering them prone to multiple interpretations and mathematical descriptions. Considering conceptual model uncertainty is thus a critical process in hydrologic uncertainty assessment. For the Death Valley Regional Flow System (DVRFS) model, developed by the U.S. Geological Survey (Belcher et al., 2004) and covering portions of southwest Nevada and southeast California, five alternative recharge models have been independently developed to date. These models are (1) the Maxey-Eakin model (Maxey-Eakin, 1949), (2 and 3) a distributed parameter watershed model with and without a runon-runoff component (Hevesi, 2003), and (4 and 5) a chloride mass balance model with two zero-recharge masks, one for alluvium and one for both alluvium and elevation (Russell and Minor, 2003). Whereas these five models are based on different methodologies for estimating recharge and have different levels of complexity, they all have been used for groundwater modeling in Nevada. The objective of our work is to evaluate recharge model uncertainty and quantify its propagation through the groundwater modeling process. We apply the recently developed Maximum Likelihood Bayesian Model Averaging (MLBMA) method (Neuman, 2003; Ye et al., 2004) and formally incorporate prior information and field measurements into the process. The DVRFS model is the numerical modeling framework and the recharge values of the five recharge models are handled by the recharge package of MODFLOW-2000. Conceptual model uncertainty is first evaluated through expert elicitation based on prior information possessed by a panel of seven experts. Their perceptions of model plausibility are quantified as prior model probabilities, which are then updated by the site measurements of head and flux through inverse modeling using the parameter estimation package in MODFLOW-2000. Posterior model probabilities of the five models are then evaluated after the updating process and used as weights in the summation of each model's mean predictions and associated predictive uncertainty. The modeling process provides the mean and variance of groundwater flux with consideration of both parametric and conceptual model uncertainty.
H13A-1317
Evaluation of Alternatives to the Southwest No-Flow Boundary Condition in the Culebra at the Waste Isolation Pilot Plant
The Culebra Dolomite within the Rustler Formation at the Waste Isolation Pilot Plant (WIPP) has been identified as a potential pathway for radionuclide migration to the environment. Water level measurements indicate a groundwater divide exists within the Culebra southwest of the WIPP site. This project evaluates the influence of the boundary condition used to represent the groundwater divide on the estimated solute travel time from the center of the WIPP repository to the site boundary. The southwest boundary condition is presently represented by a no-flow boundary extending southeast from a potash tailings pile in Nash Draw. Alternative conceptualization of this boundary is desired as the groundwater divide could be a result of infiltration where the Culebra is unconfined. In this regard, alternate boundary conceptualizations include (1) fixed heads along the structural high in the Culebra in the southwest region and (2) moving the no-flow boundary to the west in order to include recharge where the Culebra is unconfined. For each change, stochastic inverse calibration of the Culebra transmissivity fields to both steady-state heads obtained during calendar year 2000 and to a series of transient responses to various hydraulic tests over a period of 11 years is carried out. Preliminary results from 15 realizations of the base transmissivity field show that solute travel time within the WIPP site is responsive to the alternate conceptualizations of the groundwater divide within the Culebra Dolomite. This research is funded by WIPP programs administered by the Office of Environmental Management (EM) of the U.S. Department of Energy. Sandia is a multiprogram laboratory operated by Sandia Corporation, a Lockheed Martin Company, for the United States Department of Energy's National Nuclear Security Administration under contract DE-AC04-94AL85000.
H13A-1318
Assessment of Alternative Conceptual Models Using Reactive Transport Modeling with Monitoring Data
Monitoring data proved very useful in evaluating alternative conceptual models, simulating contaminant transport behavior, and reducing uncertainty. A graded approach using three alternative conceptual site models was formulated to simulate a field case of tetrachloroethene (PCE) transport and biodegradation. These models ranged from simple to complex in their representation of subsurface heterogeneities. The simplest model was a single-layer homogeneous aquifer that employed an analytical reactive transport code, BIOCHLOR (Aziz et al., 1999). Due to over-simplification of the aquifer structure, this simulation could not reproduce the monitoring data. The second model consisted of a multi-layer conceptual model, in combination with numerical modules, MODFLOW and RT3D within GMS, to simulate flow and reactive transport. Although the simulation results from the second model were comparatively better than those from the simple model, they still did not adequately reproduce the monitoring well concentrations because the geological structures were still inadequately defined. Finally, a more realistic conceptual model was formulated that incorporated heterogeneities and geologic structures identified from well logs and seismic survey data using the Petra and PetraSeis software. This conceptual model included both a major channel and a younger channel that were detected in the PCE source area. In this model, these channels control the local ground-water flow direction and provide a preferential chemical transport pathway. Simulation results using this conceptual site model proved compatible with the monitoring concentration data. This study demonstrates that the bias and uncertainty from inadequate conceptual models are much larger than those introduced from an inadequate choice of model parameter values (Neuman and Wierenga, 2003; Meyer et al., 2004; Ye et al., 2004). This case study integrated conceptual and numerical models, based on interpreted local hydrogeologic and geochemical data, with detailed monitoring plume data. It provided key insights for confirming alternative conceptual site models and assessing the performance of monitoring networks. A monitoring strategy based on this graded approach for assessing alternative conceptual models can provide the technical bases for identifying critical monitoring locations, adequate monitoring frequency, and performance indicator parameters for performance monitoring involving ground-water levels and PCE concentrations.
H13A-1319
Multiobjective Time-Dependent Optimal Experimental Design of Sampling Networks
This work addresses the application of optimal experimental design (OED) for obtaining groundwater sampling data in an efficient manner. The OED finds application here in developing designs for a complex reactive transport scenario. In this context, monitoring networks can be used for multiple purposes, such as for collecting samples that improve the accuracy of reaction parameters and for reliably monitoring contaminant transport. The nondominated sorting genetic algorithm, NSGA-II, is used to develop the Pareto surface of these objectives and their tradeoff with design cost. Costs explicitly include the designer's discount rate, a factor which can significantly influence the selection of well installation and sampling times. As the temporal domain is of interest to both design cost and reaction kinetics, the methodology explicitly addresses the time dependency of the OED problem.
H13A-1320
Robust remediation design of a groundwater model with variable hydrstratigraphic units and spatial uncertainty
Including uncertainty in pump-and-treat groundwater remediation designs has been shown to be important in developing a design that is robust and avoids the risks involved in using purely deterministic modeling techniques to predict remediation response. When there are differences in the hydrologic properties within one model, the geometry and scale of these differences can govern the importance of including the uncertainty in the remediation design. Through the use of an optimization approach with recourse, i.e. a multiscenario approach, techniques for including the uncertainty of the K fields of models with varying hydrostratigraphic regions is explored. Because this is a multiscenario approach, an effort is made to determine an approach that utilizes the fewest number of scenarios to obtain a solution that is representative of the true uncertain nature of the problem. Three sampling techniques are explored. These methods include: (1) Monte-Carlo simulations whereby each hydrostratigraphic unit is represented by a correlated but randomly sampled distribution of K values; (2) Monte-Carlo simulations whereby each hydrostratigraphic unit is represented by a single randomly sampled K value and scenarios are constructed by pairing sequential K values sampled from each field; (3) Equal-area simulations whereby each hydrostratigraphic unit is represented by a single sampled K value determined through equal area sampling.
H13A-1321
Hydrologic Scenario Uncertainty in a Comprehensive Assessment of Hydrogeologic Uncertainty
A method to jointly assess hydrogeologic conceptual model and parameter uncertainties has recently been developed based on a Maximum Likelihood implementation of Bayesian Model Averaging (MLBMA). Evidence from groundwater model post-audits suggests that errors in the projected future hydrologic conditions of a site (hydrologic scenarios) are a significant source of model predictive errors. MLBMA can be extended to include hydrologic scenario uncertainty, along with conceptual model and parameter uncertainties, in a systematic and quantitative assessment of predictive uncertainty. Like conceptual model uncertainty, scenario uncertainty is represented by a discrete set of alternative scenarios. The effect of scenario uncertainty on model predictions is quantitatively assessed by conducting an MLBMA analysis under each scenario. We demonstrate that posterior model probability is a function of the scenario only through the possible dependence of prior model probabilities on the scenario. As a result, the model likelihoods (computed from calibration results), are not a function of the scenario and do not need to be recomputed under each scenario. MLBMA results for each scenario are weighted by the scenario probability and combined to render a joint assessment of scenario, conceptual model, and parameter uncertainty. Like model probability, scenario probability represents a subjective evaluation, in this case of the plausibility of the occurrence of the specific scenario. Because the scenarios describe future conditions, the scenario probabilities represent prior estimates and cannot be updated using the (past) system state data as is used to compute posterior model probabilities. Assessment of hydrologic scenario uncertainty is illustrated using a site-specific application considering future changes in land use, dam operations, and climate. Estimation of scenario probabilities and consideration of scenario characteristics (e.g., timing, magnitude) are discussed.
H13A-1322
Analysis of Model Uncertainties to Support Risk-Based Decisions Regarding Groundwater Contamination
Model simulations are widely used in environmental management decision processes. However, there are various sources of uncertainty that commonly impact the model results. Consequently, it is crucial to account for all the possible model uncertainties that impact the model results so that they are adequately considered in the management decision process. Here we discuss an uncertainty analysis of model simulations related to a contamination site located within Los Alamos National Laboratory, NM. We describe how uncertainties are quantified and propagated through a series of coupled groundwater models and then used in a risk-based decision analysis to identify and rank alternative actions to protect the environment and water users from potential impacts of groundwater contamination from former liquid-effluent discharges. Uncertainties in the contaminant source, infiltration distribution, and transport through the unsaturated and saturated zones are analyzed using a series of alternative conceptual models and stochastic model parameters. Alternative conceptual models and uncertain model parameters are defined to encompass a large range of possible uncertainties associated with potential groundwater flow and transport based on existing data and expert knowledge about the system. In all, eight alternative conceptual models using 38 uncertain parameters were analyzed. For each conceptual model and related stochastic parameter realization, we simulate contaminant transport from the contaminant outfall to water-supply wells over the next 1000 years. Based on the simulated contaminant concentrations in the groundwater pumped by water-supply wells, we evaluate health risk for the receptors. Based on the model results, sensitivity analysis is applied to identify the parameters and conceptual model elements causing high concentrations at the water-supply wells. Decision analysis is applied to define the optimal course(s) of action, which may include clean-Up, stabilization, additional characterization, and monitoring. If additional characterization is identified as an action that can reduce risk, the sensitivity analysis yields information not only about which parameters should be better characterized (and which should not), but also to what degree the uncertainty or variability in a specific parameter should be reduced. If the uncertainty is reduced to within the defined limits through characterization, then an updated risk assessment would calculate reduced risk. The results of our analysis demonstrate that due to dilution of the contaminants in the regional aquifer and within the water-supply wells, all of the alternative conceptual models yield low, calculated risk for the receptors. The uncertainty in model predictions is affected principally by the uncertainty in conceptualization. Currently, field exploration and study at the site continues, and these new data will allow us to test whether the uncertainties included in the risk assessment are broad enough so that the obtained conclusions will not change.
H13A-1323
Optimal Experiment Design for Distributed Parameter Identification in Groundwater Modeling: Case Study Warren Subbasin, California
This paper develops an optimal experimental design procedure for distributed parameter identification in groundwater modeling using the worst-case parameter (WCP) scenario. The WCP is defined as the parameter that causes the maximum deviation in model application when its structure is simplified. Consequently, the WCP of a structure is a parameter that makes the structure most difficult to homogenize. The criterion adopted for optimal design is either to minimize the total experimental cost subject to information requirement, or to maximize a measure of the information matrix subject to budget constraints. The proposed procedure allows one to find a minimum cost design that will provide sufficient information for identifying the WCP. The concept of WCP and the optimal experimental design procedure are applied to the Warren Subbasin, California. A groundwater flow model and a solute-transport model are developed for the Warren groundwater basin. Historical observations are available for model development and calibration for the period 1956-2001. At the present time, certain parts of the basin are contaminated by nitrate with concentrations exceeding the U.S. Environmental Protection Agency (USEPA) maximum contaminant level (MCL) of 44mg/l. Hence, the objectives of this case study are to: (1) identify the complexity of flow and transport model structures, (2) evaluate the data sufficiency determined by the objectives and accuracy requirement of model application, and (3) propose a conjunctive use program which will decrease the high nitrate concentration while maintaining the water table at the desired level.
H13A-1324
System Dynamics Approach for Critical Infrastructure and Decision Support. A Model for a Potable Water System.
The Critical Infrastructure Protection / Decision Support System (CIP/DSS) project, supported by the Science and Technology Office, has been developing a risk-informed Decision Support System that provides insights for making critical infrastructure protection decisions. The system considers seventeen different Department of Homeland Security defined Critical Infrastructures (potable water system, telecommunications, public health, economics, etc.) and their primary interdependencies. These infrastructures have been modeling in one model called CIP/DSS Metropolitan Model. The modeling approach used is a system dynamics modeling approach. System dynamics modeling combines control theory and the nonlinear dynamics theory, which is defined by a set of coupled differential equations, which seeks to explain how the structure of a given system determines its behavior. In this poster we present a system dynamics model for one of the seventeen critical infrastructures, a generic metropolitan potable water system (MPWS). Three are the goals: 1) to gain a better understanding of the MPWS infrastructure; 2) to identify improvements that would help protect MPWS; and 3) to understand the consequences, interdependencies, and impacts, when perturbations occur to the system. The model represents raw water sources, the metropolitan water treatment process, storage of treated water, damage and repair to the MPWS, distribution of water, and end user demand, but does not explicitly represent the detailed network topology of an actual MPWS. The MPWS model is dependent upon inputs from the metropolitan population, energy, telecommunication, public health, and transportation models as well as the national water and transportation models. We present modeling results and sensitivity analysis indicating critical choke points, negative and positive feedback loops in the system. A general scenario is also analyzed where the potable water system responds to a generic disruption.
H13A-1325
Using Temporal Persistence to Upscale Soil Water Contents and Reduce Uncertainty
When a field plot or a small watershed is repeatedly surveyed for soil water contents, locations can often be identified where soil is consistently wetter or dryer than the average across the surveyed area. The objective of this work was to upscale water contents from point measurements to the field scale. To accomplish this, a technique was developed using temporal persistence of soil water contents to reduce uncertainty in the average water contents. 24 soil moisture multi-sensor capacitance probes were installed to monitor water content across a 6 ha area at the USDA-ARS OPE3 site in Beltsville, MD. These probes were located at depths of 10, 30, 50, 80, 120, 150, and 180 cm and were monitored every 10 min for 610 days. To quantify the temporal persistence, hourly average water content was computed for all probe readings at one depth. Then the relative water contents were computed as ratios of the individual-probe water content measurement to the average water contents at that same depth (i.e. 10 cm depth). Based on these calculations, it appears that temporal persistence of soil-water contents was well pronounced at certain probe locations. For example, the distribution of relative water content covered a narrow range (e.g., 0.8 to 1) for the location 20. Median relative water contents were used to estimate missing data or correct errant sensor readings, thus substantially reducing uncertainty in average water content across the study area. Relative water contents also enabled reducing the number of sensors needed to obtain a specified accuracy of the average water content estimates. One month of soil moisture monitoring was found sufficient to evaluate distributions of the relative water contents and thus to determine temporal persistence. Using temporal persistence was a useful means to upscale water contents and reduce uncertainty.
H13A-1326
Analysis of Uncertainty in Discharge Data and Assessing its Impact on Model Calibration
In hydrological practice, discharge data, which is considered observed, is not actually observed because continuous measurement of discharge is time consuming, costly and infeasible during high floods. Therefore, most discharge records are developed from converting the water level data to discharge data by using a stage-discharge relationship, which is also called rating curve. The discharge data is subjected to uncertainty due to a number of factors. These factors can be classified into two categories: measurement uncertainty and rating curve uncertainty. In the development of hydrological models, calibration of model parameters is based on the discharge data. Therefore, the quality of discharge data is extremely important for reducing uncertainty in model parameters. In hydrological modeling, the uncertainty in discharge measurement is either ignored or assumed error range is applied to understand the influence of error in discharge data. Therefore, the objective of the study is to find out the ranges of error in discharge data due to measurement error as well as rating curve error and to use this range in Monte Carlo framework to analyze to what extent hydrological model parameters and model performance is affected. The study area for the research is the West Rapti River Basin, which is located in the south west of Nepal. Jalkundi gauging station, the most downstream station is used as an example for the analysis of uncertainty. The discharge is measured by velocity area method and the stage is measured by staff gauge. First, the magnitude of uncertainty in discharge measurement due to the uncertainty in velocity measurement and uncertainty in cross-sectional properties measurement is computed. Then, a rating curve in the form of power equation is developed using least square optimization. The uncertainty due to rating curve is analyzed statistically. Based on the uncertainty range found from the analysis, perturbation to discharge is applied and NAM, a lumped and conceptual hydrological model is calibrated for each realization using SCE-UA method. The result of the study shows that the uncertainty in discharge for the station under study ranges from 10%-16% at 95% confidence interval. The uncertainty in discharge has significant influence on the model parameter and model performance.
H13A-1327
How Well Does Fracture Set Characterization Reduce Uncertainty in Capture Zone Size for Wells Situated in Sedimentary Bedrock Aquifers?
Regional groundwater flow models are rife with uncertainty. The three-dimensional flux vector fields must generally be inferred using inverse modelling from sparse measurements of hydraulic head, from measurements of hydraulic parameters at a scale that is miniscule in comparison to that of the domain, and from none to a very few measurements of recharge or discharge rate. Despite the inherent uncertainty in these models they are routinely used to delineate steady-state or time-of-travel capture zones for the purpose of wellhead protection. The latter are defined as the volume of the aquifer within which released particles will arrive at the well within the specified time and their delineation requires the additional step of dividing the magnitudes of the flux vectors by the assumed porosity to arrive at the ''average linear groundwater velocity'' vector field. Since the porosity is usually assumed constant over the domain one could be forgiven for thinking that the uncertainty introduced at this step is minor in comparison to the flow model calibration step. We consider this question when the porosity in question is fracture porosity in flat-lying sedimentary bedrock. We also consider whether or not the diffusive uptake of solute into the rock matrix which lies between the source and the production well reduces or enhances the uncertainty. To evaluate the uncertainty an aquifer cross section is conceptualized as an array of horizontal, randomly-spaced, parallel-plate fractures of random aperture, with adjacent horizontal fractures connected by vertical fractures again of random spacing and aperture. The source is assumed to be a continuous concentration (i.e. a dirichlet boundary condition) representing a leaking tank or a DNAPL pool, and the receptor is a fully pentrating well located in the down-gradient direction. In this context the time-of-travel capture zone is defined as the separation distance required such that the source does not contaminate the well beyond a threshold concentration within the specified time. Aquifers are simulated by drawing the random spacings and apertures from specified distributions. Predictions are made of capture zone size assuming various degrees of knowledge of these distributions, with the parameters of the horizontal fractures being estimated using simulated hydraulic tests and a maximum likelihood estimator. The uncertainty is evaluated by calculating the variance in the capture zone size estimated in multiple realizations. The results show that despite good strategies to estimate the parameters of the horizontal fractures the uncertainty in capture zone size is enormous, mostly due to the lack of available information on vertical fractures. Also, at realistic distances (less than ten kilometers) and using realistic transmissivity distributions for the horizontal fractures the uptake of solute from fractures into matrix cannot be relied upon to protect the production well from contamination.
H13A-1328
Use of Groundwater Lifetime Expectancy for the Performance Assessment of Deep Geologic Radioactive Waste Repositories.
Long-term solutions for the disposal of toxic wastes usually involve isolation of the wastes in a deep subsurface geologic environment. In the case of spent nuclear fuel, the safety of the host repository depends on two main barriers: the engineered barrier and the natural geological barrier. If radionuclide leakage occurs from the engineered barrier, the geological medium represents the ultimate barrier that is relied upon to ensure safety. Consequently, an evaluation of radionuclide travel times from the repository to the biosphere is critically important in a performance assessment analysis. In this study, we develop a travel time framework based on the concept of groundwater lifetime expectancy as a safety indicator. Lifetime expectancy characterizes the time radionuclides will spend in the subsurface after their release from the repository and prior to discharging into the biosphere. The probability density function of lifetime expectancy is computed throughout the host rock by solving the backward-in-time solute transport equation subject to a properly posed set of boundary conditions. It can then be used to define optimal repository locations. In a second step, the risk associated with selected sites can be evaluated by simulating an appropriate contaminant release history. The proposed methodology is applied in the context of a typical Canadian Shield environment. Based on a statistically-generated three-dimension network of fracture zones embedded in the granitic host rock, the sensitivity and the uncertainty of lifetime expectancy to the hydraulic and dispersive properties of the fracture network, including the impact of conditioning via their surface expressions, is computed in order to demonstrate the utility of the methodology.
H13A-1329
Numerical Volumetric Analysis of Spatially Dependent Transmissivity and Storativity in Heterogeneous Aquifers
Aquifer transmissivity (T) and storativity (S) values control the rate and areal extent of propagation of the cone of depression from a pumped well. It has been documented recently that transmissivity and storativity, reflect the geometric and arithmetic means, respectively, of the area contacted by the cone of depression. However, these findings do not reflect a volumetric evaluation of the cone of depression within the heterogenities. Analysis of spatial, volumetric variations within the cone of depression expressed at the potentiometric surface, offers a more general solution to evaluate the meaning of T and S values. Log-Normal and normal distributions of hydraulic conductivity as block heterogeneities were established within model domains for simulated aquifer tests. By analyzing the volumetric evolution of the cone of depression observed in the potentiometric surface, we are able to illustrate the averaging of transmissivity as a function of time, and distance from the pumping well for the entire affected aquifer. Volumetric analysis of simulated aquifer tests show an exponential decrease in the arithmetic, harmonic, and geometric weighted mean transmissivity within the evolving cone of depression through time, approaching steady, basin-wide averages. Transmissivity estimates derived for single observation wells by conventional testing methods (i.e., Theis, 1935; Cooper and Jacob, 1946) are found to increase with increasing radial distance from the pumping well. However, when the same observation well drawdown data are plotted together with, and constrained by, the drawdown curve for the pumping well, a family of drawdown curves is derived that yields transmissivity values that are consistent with the volumetric, weighted, mean transmissivity values calculated for the entire cone of depression for specific periods of time (i.e., areal extent).
H13A-1330
Nonpoint Source Solute Transport Normal to Aquifer Bedding in Heterogeneous, Markov Chain Random Fields
Facies-based geostatistical models have become important tools for the stochastic analysis of flow and transport processes in heterogeneous aquifers. However, little is known about the dependency of these processes on the parameters of facies- based geostatistical models. This study examines the nonpoint source solute transport normal to the major bedding plane in the presence of interconnected high conductivity (coarse- textured) facies in the aquifer medium and the dependence of the transport behavior upon the parameters of the constitutive facies model. A facies-based Markov chain geostatistical model is used to quantify the spatial variability of the aquifer system hydrostratigraphy. It is integrated with a groundwater flow model and a random walk particle transport model to estimate the solute travel time probability distribution functions (pdfs) for solute flux from the water table to the bottom boundary (production horizon) of the aquifer. The cases examined include, two-, three-, and four-facies models with horizontal to vertical facies mean length anisotropy ratios, ek, from 25:1 to 300:1, and with a wide range of facies volume proportions (e.g, from 5% to 95% coarse textured facies). Predictions of travel time pdfs are found to be significantly affected by the number of hydrostratigraphic facies identified in the aquifer, the proportions of coarse-textured sediments, the mean length of the facies (particularly the ratio of length to thickness of coarse materials), and - to a lesser degree - the juxtapositional preference among the hydrostratigraphic facies. In transport normal to the sedimentary bedding plane, travel time pdfs are not log- normally distributed as is often assumed. Also, macrodispersive behavior (variance of the travel time pdf) was found to not be a unique function of the conductivity variance. The skewness of the travel time pdf varied from negatively skewed to strongly positively skewed within the parameter range examined. We also show that the Markov chain approach may give significantly different travel time pdfs when compared to the more commonly used Gaussian random field approach even though the first and second order moments in the geostatistical distribution of the lnK field are identical. The choice of the appropriate geostatistical model is therefore critical in the assessment of nonpoint source transport.
H13A-1331
Parameter Sensitivity Analyses For A Large-Scale Unsaturated Flow Model
Field investigations for the Yucca Mountain site have shown that large variabilities exist in flow parameters over the spatial domain of the mountain. Even though considerable progress has been made in model development for this area, uncertainty associated with the site-scale unsaturated zone (UZ) flow model input parameters has not been investigated in depth. In this study, parameter sensitivity analyses have been conducted using the site-scale 3-D UZ flow model. The sensitivity analyses are intended to evaluate the effects of uncertainties in hydrologic parameters on UZ flow and contaminant transport. Sensitivity analyses are carried out using the parameter uncertainties for fracture and matrix permeabilities and van Genuchten alphas, which are the most sensitive parameters affecting UZ flow simulations. These sensitivity simulations are performed using the UZ flow model by incrementing or decrementing a given parameter by one standard deviation, a modification performed for each parameter in all the hydrogeologic units/layers as well as faults. The parameter variation results in a total of eight parameter sets to account for the uncertainties of the four hydrological parameters. Modeling results for the eight flow simulation sensitivity cases have been compared with observed borehole matrix liquid saturation, matrix water potential, and perched water data. In addition, the eight simulations are also compared with simulation results from the base-case parameter set. The effects of parameter uncertainties on the flow fields are discussed in detail through the comparison of results.